Particle-based Window Rotation and Scaling Scheme for Real-time Hand Recognition and Tracking

Abstract

In this paper, we propose a Particle-based Window Rotation and Scaling (PWRS) algorithm, which is a multi-stage system that can perceive hand size and rotating angle using a single camera. There are three stages of operation in the PWRS scheme, including window-locating stage, window-scaling stage and window-rotating stage, that are adopted to recognize the location, size and angle of hand motion, respectively. Each stage employs histogram of oriented gradients, support vector machine, and particle filter to detect and track the hand window. Unlike traditional multi-stage system which requires to detect and then remove non-hand regions case-by-case, the PWRS scheme can preserve similar characteristics at each stage and predict the results propagated from other stages by cross-stage propagation method. This architecture allows each stage to focus on its own target characteristics so as to detect and track in a Wdiversity-reduced space. Experimental results show that the proposed PWRS algorithm can effectively provide satisfactory hand motion recognition and tracking for real-time applications.